In this letter, we investigate the problem of dynamic spectrum access for small cell networks, using a graphical game approach. Compared with existing studies, we take the features of different cell loads and local interference relationship into account. It is proved that the formulated spectrum access game is an exact potential game with the aggregate interference level as the potential function, and Nash equilibrium (NE) of the game corresponds to the global or local optima of the original optimization problem. A lower bound of the achievable aggregate interference level is rigorously derived. Finally, we propose an autonomous best response learning algorithm to converge towards its NE. It is shown that the proposed game-theoretic solution converges rapidly and its achievable performance is close to the optimum solution.
6 pages, 5 figures, Submitted to IEEE Transactions on Vehicular Technology as a correspondence (under third round review)